Utilização de random forest para estimar o volume de madeira em pátios de estocagem

• Autor(a) principal: Lobato, Lucas José Teodoro
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Espírito Santo (riUfes)
• Texto Completo: http://repositorio.ufes.br/handle/10/12500
• topic: Recursos Florestais e Engenharia Florestal; Processamento de imagens; Estimativa do volume de madeira; Aprendizado de máquinas; Uso de diferentes distâncias na fotografia das pilhas montadas; Inclinação do smartphone

The Brazilian forestry sector has stood out in relation to other sectors due to investments, use of new technologies and favorable soil and climate conditions for the development of forestry species. With this growth, forestry companies have been seeking methodologies that make it possible to estimate the volume of wood more quickly and with low operating costs. Given the difficulty and errors made when using traditional methods to measure the volume of wood in storage yards, the use of digital images appears as a viable and quick alternative to estimate the volume of wood. Despite the excellent results using digital images to estimate the volume of wood contained in piles, there is still limited work evaluating the accuracy of this methodology using images obtained from a smartphone, as well as the correct way to obtain these images. Due to these factors, the objective of this work was to estimate the volume of Eucalyptus wood in a storage yard with digital images obtained with a smartphone using Random Forest. These images were taken at different distances and inclinations. The work was carried out in a storage yard of a sawmill located in the municipality of Ibitirama. 30 piles of wood were assembled, and digital images were obtained of both sides of the piles, at a distance of 1.5; 2.5; 4.5 and 6.5 meters. Furthermore, images were obtained with the smartphone tilted horizontally (10°, 20° and 30°) and vertically (5°, 7° and 10°) for distances of 2.5; 4.5 and 6.5 meters. After obtaining the digital images, a machine learning model (Random Forest) was trained to identify the wood class and reference object of the images using the R programming environment. After processing the images, it was assessed that the images taken 6.5 and 4.5 meters from the stack had greater accuracy (2.30% and 3.40%) in the estimated volume compared to the images obtained at 2.5 and 1.5 m (9.29% and 17, 61%). Furthermore, it was observed that the greater the inclination of the smartphone, the greater the error made when estimating the volume of wood contained in the assembled piles. According to the data obtained, it is concluded that: (i) the use of digital images obtained with a smartphone was accurate to estimate the volume of wood located in storage yards, (ii) images obtained further away from the wood piles showed greater accuracy compared to the closest ones, (iii) there was a significant influence on the estimation of the volume of wood, due to the inclination of the smartphone, which is less accentuated in images obtained further away from the pile of wood, (iv) the vertical displacement of the smartphone , had a greater influence on the estimation of the volume of wood, due to the increase in the sectional area of the wood, caused by the presence of the side and upper part of the logs in these images.